Let ocfs2 use the new truncate sequence. The changes include:
1. Move inode_change_ok into cluster lock and remove inode_newsize_ok.
2. Use truncate_setsize directly since we don't implement our
own ->truncate and what we need is "update i_size and
truncate_pagecache" which truncate_setsize now does.
3. Change some i_size_read to inode->i_size in ocfs2_setattr
since we have i_muext held.
4. For direct write, ocfs2 actually don't allow write to pass
i_size(see ocfs2_prepare_inode_for_write), so we don't have
a chance to increase i_size. So remove the bogus check.
Cc: Joel Becker <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Nick Piggin <[email protected]>
Signed-off-by: Tao Ma <[email protected]>
---
fs/ocfs2/file.c | 45 ++++++++++-----------------------------------
1 files changed, 10 insertions(+), 35 deletions(-)
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 1fb0985..6e4685e 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -960,10 +960,6 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
return 0;
}
- status = inode_change_ok(inode, attr);
- if (status)
- return status;
-
if (is_quota_modification(inode, attr))
dquot_initialize(inode);
size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
@@ -982,12 +978,12 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
goto bail_unlock_rw;
}
- if (size_change && attr->ia_size != i_size_read(inode)) {
- status = inode_newsize_ok(inode, attr->ia_size);
- if (status)
- goto bail_unlock;
+ status = inode_change_ok(inode, attr);
+ if (status)
+ goto bail_unlock;
- if (i_size_read(inode) > attr->ia_size) {
+ if (size_change && attr->ia_size != inode->i_size) {
+ if (inode->i_size > attr->ia_size) {
if (ocfs2_should_order_data(inode)) {
status = ocfs2_begin_ordered_truncate(inode,
attr->ia_size);
@@ -1052,22 +1048,12 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
}
/*
- * This will intentionally not wind up calling truncate_setsize(),
- * since all the work for a size change has been done above.
- * Otherwise, we could get into problems with truncate as
- * ip_alloc_sem is used there to protect against i_size
- * changes.
- *
- * XXX: this means the conditional below can probably be removed.
+ * Since all the work for a size change has been done above.
+ * Call truncate_setsize directly to change size and truncate
+ * pagecache.
*/
- if ((attr->ia_valid & ATTR_SIZE) &&
- attr->ia_size != i_size_read(inode)) {
- status = vmtruncate(inode, attr->ia_size);
- if (status) {
- mlog_errno(status);
- goto bail_commit;
- }
- }
+ if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size != inode->i_size)
+ truncate_setsize(inode, attr->ia_size);
setattr_copy(inode, attr);
mark_inode_dirty(inode);
@@ -2122,17 +2108,6 @@ relock:
written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
ppos, count, ocount);
if (written < 0) {
- /*
- * direct write may have instantiated a few
- * blocks outside i_size. Trim these off again.
- * Don't need i_size_read because we hold i_mutex.
- *
- * XXX(truncate): this looks buggy because ocfs2 did not
- * actually implement ->truncate. Take a look at
- * the new truncate sequence and update this accordingly
- */
- if (*ppos + count > inode->i_size)
- truncate_setsize(inode, inode->i_size);
ret = written;
goto out_dio;
}
--
1.5.5
On Fri, Jun 11, 2010 at 11:27:49AM +0800, Tao Ma wrote:
> Let ocfs2 use the new truncate sequence. The changes include:
> 1. Move inode_change_ok into cluster lock and remove inode_newsize_ok.
> 2. Use truncate_setsize directly since we don't implement our
> own ->truncate and what we need is "update i_size and
> truncate_pagecache" which truncate_setsize now does.
> 3. Change some i_size_read to inode->i_size in ocfs2_setattr
> since we have i_muext held.
> 4. For direct write, ocfs2 actually don't allow write to pass
> i_size(see ocfs2_prepare_inode_for_write), so we don't have
> a chance to increase i_size. So remove the bogus check.
>
> Cc: Joel Becker <[email protected]>
> Cc: Christoph Hellwig <[email protected]>
> Cc: Nick Piggin <[email protected]>
The patch looks fine to me. I really appreciate you working on truncate,
thanks.
This looks correct, but still has the second if ATTR_SIZE block that
I commented on last time. I'd really prefer if the filesystems could
move the truncate handling into a single conditional to simplify
auditing for it and possibly splitting it out into a separate method
later.
And btw, the S_ISREG check which you only have on the first ATTR_SIZE
check is superflous, the VFS only does ATTR_SIZE calls on regular files.
Hi Christoph,
On 06/11/2010 02:56 PM, Christoph Hellwig wrote:
> This looks correct, but still has the second if ATTR_SIZE block that
> I commented on last time. I'd really prefer if the filesystems could
> move the truncate handling into a single conditional to simplify
> auditing for it and possibly splitting it out into a separate method
> later.
oh, that would be much work for ocfs2 to do from my perspective. So I
would really want to leave it as-is and I have add it to my to-do list.
>
> And btw, the S_ISREG check which you only have on the first ATTR_SIZE
> check is superflous, the VFS only does ATTR_SIZE calls on regular files.
yeah, I can remove it.
Regards,
Tao
On Fri, Jun 11, 2010 at 03:16:13PM +0800, Tao Ma wrote:
> Hi Christoph,
>
> On 06/11/2010 02:56 PM, Christoph Hellwig wrote:
> >This looks correct, but still has the second if ATTR_SIZE block that
> >I commented on last time. I'd really prefer if the filesystems could
> >move the truncate handling into a single conditional to simplify
> >auditing for it and possibly splitting it out into a separate method
> >later.
> oh, that would be much work for ocfs2 to do from my perspective. So I
> would really want to leave it as-is and I have add it to my to-do list.
Oh right, you start a new transaction there. Sorry, ignore my request
and keep it as it is for now. I don't think doing truncatate in separate
transactions actually is correct, though but that's no change with this
patch.
On Fri, Jun 11, 2010 at 09:19:45AM +0200, Christoph Hellwig wrote:
> On Fri, Jun 11, 2010 at 03:16:13PM +0800, Tao Ma wrote:
> > oh, that would be much work for ocfs2 to do from my perspective. So I
> > would really want to leave it as-is and I have add it to my to-do list.
>
> Oh right, you start a new transaction there. Sorry, ignore my request
> and keep it as it is for now. I don't think doing truncatate in separate
> transactions actually is correct, though but that's no change with this
> patch.
Christoph,
You're missing the part where actual truncate (reduce i_size)
sets i_size in ocfs2_truncate_file(). So this later code doesn't get
triggered for the truncate case. It exists for the extend case, where
we extend the allocation in multiple clean transactions, then finally
set i_size in a final transaction.
Joel
--
"When choosing between two evils, I always like to try the one
I've never tried before."
- Mae West
Joel Becker
Principal Software Developer
Oracle
E-mail: [email protected]
Phone: (650) 506-8127
On Fri, Jun 11, 2010 at 12:50:20AM -0700, Joel Becker wrote:
> On Fri, Jun 11, 2010 at 09:19:45AM +0200, Christoph Hellwig wrote:
> > On Fri, Jun 11, 2010 at 03:16:13PM +0800, Tao Ma wrote:
> > > oh, that would be much work for ocfs2 to do from my perspective. So I
> > > would really want to leave it as-is and I have add it to my to-do list.
> >
> > Oh right, you start a new transaction there. Sorry, ignore my request
> > and keep it as it is for now. I don't think doing truncatate in separate
> > transactions actually is correct, though but that's no change with this
> > patch.
>
> Christoph,
> You're missing the part where actual truncate (reduce i_size)
> sets i_size in ocfs2_truncate_file(). So this later code doesn't get
> triggered for the truncate case. It exists for the extend case, where
> we extend the allocation in multiple clean transactions, then finally
> set i_size in a final transaction.
Actually, ocfs2_extend_file() appears to handle it too. I don't
think it used to - there were places that had issues with
commit_write()'s update of i_size, etc. But that's ancient history. I
wonder if Mark knows.
joel
--
Dort wo man B?cher verbrennt, verbrennt man am Ende auch Mensch.
(Wherever they burn books, they will also end up burning people.)
- Heinrich Heine
Joel Becker
Principal Software Developer
Oracle
E-mail: [email protected]
Phone: (650) 506-8127